Modeling the Enablers of Consumers’ E-Shopping Behavior: A Multi-Analytic Approach

Abstract

The evolution of e-commerce amid the positive growth forecast of the e-commerce market has sparked scholarly interest in e-shopping antecedents to better understand customer behavior and ensure sustainable e-shopping services. The purpose of this study is to investigate the relationship between the enablers of customers’ e-shopping intention and e-shopping behavior in the post-pandemic period. Personal innovativeness, service quality, perceived risk, and trust were incorporated into the Unified Theory of Technology Acceptance and Usage (UTAUT) original framework and UTAUT 2 in this study. To explore the relationship among the study variables, data were collected from 420 shoppers via an online survey using a convenience sampling technique. The obtained data were analyzed using a multi-analytic approach, such as structural equation modeling and artificial neural networks (SEM-ANN). The empirical findings showed that trust, habit, and e-shopping intention significantly influence consumers’ e-shopping behavior. Furthermore, the results indicated that personal innovativeness, facilitating conditions, performance expectancy, habit, effort expectancy, perceived risk, price value, hedonic motivation, service quality, and trust were all significantly linked to e-shopping intention. The study revealed that effort expectancy acts as a mediator between service quality and e-shopping behavior. This research provides valuable insights into e-shopping behavior in developing countries during the post-pandemic era. By providing a more comprehensive and accurate understanding of the factors that influence e-shopping behavior, hybrid SEM-ANN analysis can help managers and policymakers arrive at better-informed decisions to promote and encourage e-shopping.

1. Introduction

Recently, the rapid growth of e-commerce provides new models or frameworks for online businesses worldwide [1]. It is forecasted that the e-commerce market will grow significantly from $24 billion in 2019 to $98 billion in 2024 [2]. The rapid expansion of e-businesses has equally birthed a new model of distribution services, whereby individual items, as opposed to bulk items, are shipped to retail stores or directly to customers [3]. This distribution process, though necessary for businesses to remain competitive, may be unsustainable for businesses with a small customer base. In other words, the distribution channel (i.e., delivery vans) will be operating sub-optimally by traveling a long distance only to deliver a few goods [4]. Therefore, for e-shopping businesses to be profitable, they must be operated on a large scale by growing their customer base. This underscores the need to understand factors promoting the use of e-shopping services among consumers. Although multiple scholars [5,6,7] have proposed several e-shopping antecedents, literature works on the topic are still nascent. Hence, a thorough investigation is necessary to help us better understand customers’ e-shopping behavior. The recent COVID-19 pandemic, which resulted in the enforcement of lockdowns in many countries, has also given impetus to the investigation of online shopping determinants, as more and more people were forced to use online shopping services during the outbreak [8,9]. However, whether or not the so-called “forced adopters of online shopping” will continue utilizing online retail services during the post-pandemic era remain unclear. Therefore, understanding the determinants of consumers’ online-shopping behavior is crucial to ensure the sustainability of online retail services. Martín et al. [10] found that the quantity of e-shopping articles has increased over time. Still, there are some gaps on the topic of the determinants of consumerism. In addition, the majority of the study conducted to this point has been in the USA, the UK, India, and China, with developing countries, such as Bangladesh, receiving insufficient attention.

In the realm of online buying research, scholars commonly rely on theories from the field of Information Systems, such as the Theory of Reasoned Action (TRA), the Theory of Planned Behavior (TPB), and the Technology Acceptance Model (TAM), in order to gain a deeper understanding of individuals’ reactions to new technologies in the workplace. In other words, these theories explain how people accept and use new technologies solely from a technological perspective [11]. As per Childers et al. [12], a perspective that is focused on technology is likely to be wrong in a customer setting. This is true for products that have significant hedonic features and services that are goal-driven, e.g., online shopping. Other aspects, such as the individual’s innate motivations, may even be more important than the technological characteristics for understanding consumers’ online-buying intention. Klepek and Bauerová [13] conducted a contemporary study in the Czech Republic and observed that hedonic factors are essential for a thorough comprehension of non-buyers’ hesitations toward online buying.

Existing studies have revealed that aspects of Service Quality (SQ) influence consumers’ product/service evaluations. SQ is the customers’ belief that the systems or services they are using are reliable, secure, and trustworthy. According to scholars [14], the primary determinants of consumers’ online-buying decisions are aspects of web design, including information, photos, video demos, and graphics. In this highly competitive environment, the enhancement of these aspects is essential for e-businesses to attract more customers and ultimately increase their market share. Researchers [15] have underscored the necessity of interaction and personalization on e-commerce websites, as they have a substantial impact on the SQ and customer experience [15]. Khan et al. [16] emphasized the influence of technology and innovation on consumer behavior, while Nyagadza [17] reviewed sustainable growth strategies for digital marketing transformation. Lu et al. [18] examined how the affordances of e-commerce live streaming affect consumers’ gift-giving behavior and purchasing intentions. Likewise, existing research [19] used satisfaction as a mediator between SQ and behavioral intention. However, no research has studied the mediation role of effort expectancy, except Amjad-ur-Rehman et al. [20] who proved that the effort-free feature of a system strengthens the linkage between SQ and the behavioral intention of the user. This relationship could, however, be retested or verified from other countries’ perspectives, as it is under-researched.

Despite being a driver of online behavior, the Perceived Risk (PR) factor is not incorporated in the widely recognized technology adoption models, such as TAM and UTAUT [21]. Meanwhile, perceived risk is very relevant to e-commerce, considering that customers cannot touch or feel products they ordered online, thus raising their concerns about the safety and quality of online shopping services. Only a few scholars in the area of online shopping have used PR in their assessments. Hansen [22] defined PR as “the degree to which a consumer believes it is risky (e.g., poor payment security and untrustworthy online stores) to use the web for necessary shopping or that adverse consequences are conceivable.” Hansen discovered that consumers in Sweden had a negative attitude toward purchasing food online and the intention to do so and that this attitude is adversely associated with perceived risk [22]. In another study conducted in the US, the author [23] pointed out that the PR level does not vary much among people who do not shop online, people who buy products other than groceries online, and people who use online grocery websites. This underscores the need to test perceived risk in other settings using the UTAUT2 model.

In this paper, we relied on the UTAUT2, a “meta-model” proposed by Venkatesh et al. [24]. The main benefit of the model is that it takes into account consumer-context factors, such as hedonic motivation (HM), price value (PV), and habit. Three context-specific components are included in the UTAUT2 model to distinguish between online purchasing and traditional shopping: perceived risk, SQ, and personal innovativeness. Therefore, the primary focus of our investigation is to determine whether or not the UTAUT2 model, in its expanded form, applies to the phenomenon of online purchasing. To be more specific, our goal is to determine which of the suggested new structures, if any at all, contribute to the creation of additional value. According to Mortimer et al. [25], online purchasing literature has so far concentrated on behavioral intention (BI); however, we focus on investigating the actual usage. By accomplishing the aforementioned objectives, the research conducted will provide various contributions to the existing body of knowledge in the field of e-commerce. As an instance, this research will prove the relevance of UTAUT2 in the settings of online shopping in a developing country. In addition, this study is unique in that a more thorough and context-based framework of UTAUT2 has been built and tested.

The remainder of the paper is arranged as follows: Section 2 of this study highlighted the theoretical foundation for the suggested causal model’s linkages. In Section 3, we discuss the research technique that was applied, and in Section 4, we explain and evaluate the findings obtained by applying the model to the collected samples. The research concludes with the most important theoretical and practical implications, along with the research’s constraints.

2. Literature Review and Hypothesis Development

2.1. Unified Theory of Acceptance and Use of Technology (UTAUT)

The Unified Theory of Acceptance and Utilize of Technology (UTAUT) is a model that was developed by researchers [24] to explain how users intend to utilize an information system and apply it [26]. Since the introduction of UTAUT, it has been subjected to empirical testing and has been validated as an advanced theory that can describe the behavior of online shoppers [27]. This model has four factors: social influence, performance expectancy, facilitating conditions, and efforts expectancy. User intent and behavior are directly predicted by the first three, whereas user behavior is solely predicted by the fourth. It is believed that experience, age, gender, and voluntariness of use each play a role in moderating the effect of the four key components on people’s intentions and behavior. The UTAUT is the result of an exhaustive literature review and synthesis of eight distinct models, which include the TRA, TAM, the motivational model (MM), a hybrid of the TPB and TAM, the model of PC usage, the innovation diffusion theory (IDT), and the social cognitive theory (SCT) [24,26]. In a longitudinal investigation, researchers found that UTAUT was responsible for the differences in usage intention and actual use [26].

Table 1. Empirical literature on e-shopping.

Source	Aspect/Country	Methodology	Constructs Tested
[28]	e-shopping adoption/Pakistan	AMOS/SEM/ UTAUT/356	Social influence, offline brand trust, Facilitating condition, performance expectancy, service quality, and effort expectancy,
[29]	Predictors of online shopping/India	AMOS/SEM/ UTAUT 2/424	Facilitating condition, Effort expectancy, social influence, hedonic motivation, reverse logistics, performance expectancy, habit, price value, Payment on delivery mode, and social media.
[30]	Online shopping behavior/Australia and USA	AMOS/SEM/ Generational Cohort Theory/745	Escapism motive, Attitude, Social motive, value motive, online shopping facility, and information search
[31]	E-shopping behavior/Pakistan	AMOS/SEM/TPB/439	subjective norms, E-shopping attitude, e-shopping intention, website trust, and behavior.
[32]	E-grocery shopping/ Belgium	STATA/ Regression/UTAUT2/ 560	Trust, Effort expectancy, service quality, facilitating condition, performance expectancy, social influence, perceived risk, price value, innovativeness, perceived time pressure, perceived shopping enjoyment, hedonic motivation, and Habit
[33]	E-commerce adoption/Spain and Colombia	AMOS/SEM/ TAM/584	Buying impulse, compatibility, subjective norms, self-efficacy, perceived ease of use, perceived behavioral control, and perceived usefulness and attitude.
[34]	Shopping intention on foreign websites/ Taiwan	PLS-SEM/Attachment theory, consumer perceived value-based model, signaling theory, /678	National integrity, legal structure, website policy, past transactions, website design quality, vendor reputation, attachment avoidance, attachment anxiety, personal attachment avoidance, anxiety, return cost, waiting cost, price competitiveness, communication cost, perceived value, and product uniqueness and trustworthiness.
[35]	Purchase intention in cross-border E-commerce	LISREL/SEM/ Commitment-involvement theory/473	Product description, awareness, enduring involvement, situational involvement, benevolence, integrity, ability, and perceived trust
[36]	Online purchase intention/Malaysia	SPSS/ Regression/ 350	Security risk, financial risk, social risk, product risk, psychological risk, and time risk
[37]	Mobile apps-based shopping/India	AMOS/SEM/ UTAUT2 350	Social influence, facilitating condition, deal proneness effort, price value, expectancy, hedonic motivation, performance expectancy, and habit.

shows the existing empirical investigation on e-shopping. The conceptual framework is exhibited in the Figure 1.

2.1.1. Facilitating Conditions (FC)

In the realm of online purchasing, the term “facilitating conditions” pertains to customers’ perception of a company’s technological infrastructure and its ability to support their use of technology [26]. Specifically, it relates to the level to which a customer feels that they possess the required tools and resources, such as computers and websites, to partake in online shopping. This notion combines facilities, behavioral control, and compatibility. It is possible to measure the quality of a company or technology by assessing the impediments that may limit its use. Previous studies [28,32] have shown that FC has a significant connection with intentions. FC indicates the user’s perception that the technology they want to utilize has adequate infrastructure to support its use [24]. The necessary techno-infrastructural support for using technology is usually put under FC since it affects both what the user wants and what they do [24,26]. Tandon [29] found that FC improves behavioral intention. Facilitating conditions are seen as an indispensable factor in customer mHealth adoption [38,39]. Thus, we proposed the hypothesis below:

Hypothesis (H1). 

FC is positively linked to ESI.

2.1.2. Effort Expectancy (EE)

Individuals are more likely to accept a novel technology if it is easy to operate, according to Venkatesh [26]. EE is, therefore, denoted as the level of ease with which one can adopt a system or technology. Because many customers think of online purchasing as effortless, it is possible to measure things, such as perceived usefulness, complexity, and simplicity of use, based on how much work the buyer expects to put in. Plouffe et al. [40] pointed out that these ideas are very similar and are measured using the same scales. According to Shaikh et al. [41], EE affects a user’s desire to adopt mobile banking technology. As per Faaeq et al. [42], EE is linked with behavioral intention. Researchers [43,44,45] discovered that EE was a good measure of behavioral intention. Consequently, we formulated the following proposition:

Hypothesis (H2). 

EE is significantly related to e-shopping intention.

2.1.3. Social Influence

Individuals’ outlooks on new technology and systems are impacted by the social circles in which they participate [26]. In an e-shopping context, important people (friends/colleagues) shape customers’ perceptions. The complexities of Social Influence (SI) necessitate a wide range of auxiliary influences. SI can influence a person’s actions via three distinct channels: identification, compliance, and internalization [46]. Compliance with obligatory conditions is influenced directly by SI through these mechanisms. According to the findings of several research articles, SI plays a multidimensional role in the adaptation of novel technologies [47,48]. It is the extent to which individuals think that close people endorse their actions or technology selections [49,50]. In particular, SI has been shown to be a good predictor of health-related technology, mobile diet apps [51], wireless devices, and smartphone adoption [52]. Consequently, we posited the hypothesis below:

Hypothesis (H3). 

SI is significantly connected to ESI.

2.1.4. Performance Expectancy (PE)

In accordance with the foundational idea of perceived ease of use, we viewed PE as the extent to which persons believe e-shopping will assist them with a specific task [53]. Moreover, PE relates to the conviction that technology can facilitate the desired improvements in task execution. The extent to which a customer trusts that online purchasing can aid them in achieving a specific level of performance is referred to as PE. The PE concept encompasses five sub-components: expected outcomes, perceived usefulness, extrinsic motivation, relative advantage, and job fit. People tend to express interest in technology since they expect that technology will help them perform better and become more productive and efficient. According to past research, the intention to use healthcare technology was highly connected with PE [44,54]. Furthermore, as noted by Gu and Liu [55], the inclination to pay for QandA platforms is directly linked to the degree of personal equity. Consumers are more likely to buy something if they expect it to perform better than it currently does. As a result, we advance the following postulations:

Hypothesis (H4). 

PE and e-shopping intention are significantly and positively linked.

Hypothesis (H5). 

PE and e-shopping behavior are significantly and positively linked.

2.2. The Extension of UTAUT

While reviewing the existing literature, it was found that the UTAUT does not cover all theories. Scholars have broadened, integrated, and implemented the UTAUT in diverse business and technological contexts. For instance, researchers used it to investigate the behavior of mobile users (customers) in terms of their application of technology and acceptance of new technologies [56]. Hong and Cho [57] utilized the UTAUT to evaluate the technology-use behavior of operational, medium, and senior-level staff. The authors also evaluated people’s e-government usage [24]. Although the UTAUT can be used in many different ways, its extensions and replications are necessary for figuring out how well technology is accepted and for pushing its theoretical limits. Accordingly, scholars have produced an enhanced form of the theory (UTAUT2) by relying on the literature relating to IT acceptance and usage behavior [24]. E-shopping behavior can be better understood using the UTAUT2, which identifies more significant precursors and predictors. The UTAUT2 introduced hedonic price, habit and incentive, which promote better understanding of the connection between the facilitating conditions and the behavioral intent. New external, endogenous, moderation, and outcome mechanisms were proposed by Venkatesh et al. [58], and could be incorporated into the UTAUT. Therefore, this research also extends UTAUT by introducing one exogenous mechanism (e-shopping service quality) and one mediator variable (effort expectancy) between endogenous factors and the intention to buy online.

2.2.1. Hedonic Motivation (HM)

People are motivated both externally and internally, according to a study in psychology [59]. In the field of online buying, researchers have embraced this approach and studied both functional (utilitarian) and non-functional (hedonic) incentives, such as performance expectancy. Venkatesh et al. [24] denote HM as “the enjoyment or pleasure obtained from utilizing a technology”. In the research on e-commerce, there are not many papers that look at the role of perceived enjoyment or HM. According to Human et al. [60], there are proofs to suggest that HM has a positive effect on the BI of customers in Mauritius. In a study conducted in Malaysia [61], a favorable effect of perceived enjoyment was observed on customers’ attitudes toward online grocery shopping. Driediger and Bhatiasevi [62] concluded, based on their study in Thailand, that perceived enjoyment affects BI through PU and PEU. Ramus and Nielsen’s [63] qualitative research suggests that HM may stem from consumers’ desire to shop without packed aisles, long check-out lines, and demanding children. According to those who have used online shopping services, one of the most enjoyable aspects of the process is simply looking around at the various options. Thus, we propose the following hypothesis:

Hypothesis (H6). 

HM is positively linked to ESI.

2.2.2. Price Value (PV)

In contrast to corporate settings, consumers are frequently required to pay for services, which influences their acceptance behavior. Studies have shown that the price of delivery influences customers’ choice of “last mile” delivery modes. Additionally, it has been discovered that other aspects, including delivery mode, the window of delivery time, time spent traveling to a grocery store, and the possibility of return, are equally important determinants of BI [64,65]. According to marketing research, people evaluate the ‘perceived value of products and services by comparing their prices to the quality they receive. PV is positive in the UTAUT2 paradigm and increases the BI to use or reuse a service if its perceived advantages (which can differ among users) are larger than the monetary costs. A number of studies have shown that this relationship holds in a variety of contexts, including mobile Internet [24], electronic learning [66], m-health [39], and e-commerce [67,68,69]. In their research in Mauritius, Human et al. [61] found the existence of a direct positive impact of PV on the purchase intentions of potential adopters. In addition, the qualitative research conducted by Droogenbroeck and Van Hove [32] revealed that people considered time saving as well as greater convenience more important than the cost of a service. As a result, we make the following proposition:

Hypothesis (H7). 

PV positively affects the e-shopping intention.

2.2.3. Habit (HB)

A habit is a behavior that is conducted over and over again and usually happens without much deliberate effort or thought. Habits are patterns of behavior that can be hard to break or change because they are learned through repetition and reinforcement. For instance, if someone always looks at online stores when they have spare time, they might be more inclined to do e-shopping even if they do not intend to buy anything. In a similar way, if someone always checks their email for deals from online shops, they might be more inclined to purchase something when they get a good deal. Studies have indicated a significant correlation between habit and behavioral intention, as well as between habit and real-world usage, in a number of contexts, including mobile Internet [24], online shopping [70], e-learning [66], mobile TV [71], and mobile banking [72]. As a predictor variable, HB was found to have a favorable impact on potential adopters’ BI [60]. This is the only study in the e-shopping literature to incorporate HB as a predictor variable, and as a result, we believe that:

Hypothesis (H8). 

HB positively affects e-shopping intention.

Hypothesis (H9). 

HB positively affects e-shopping behavior.

2.2.4. E-Shopping Intention (ESI)

Scholars have been very interested in understanding how people act when they shop online. There are many different theoretical models connected to online purchases that have been used to assess both the customers’ intention and their actual behavior. According to Venkatesh et al. [24], there exists a relationship between an individual’s behavioral intention and subsequent behavior. Thus, an analysis of the customers’ intents could reveal relevant clues about their behavior. Several researchers, including [73] and Rezaei [74], have also demonstrated a direct correlation between intention and behavior. Awwad and Al-Majali [75] found that intention is a significant component of actual consumer use and serves as a substitute for actual behavior. This is also corroborated by other studies in the contexts of mobile apps [76] and 3G technology adoption [77]. So, we hypothesize that:

Hypothesis (H10). 

ESI is positively linked with ESB.

2.3. Context-Specific Factors

2.3.1. Service Quality (SQ)

Santos [78] asserts that e-service quality is “consumers” overall evaluation and judgment of the excellence and quality of e-service offerings in the virtual marketplace. Consumers may reconsider their decision to purchase food online if the transaction is plagued by problems, such as deliveries that are either delayed or incomplete, poor packing or picking of goods, unpleasant deliveries, or unacceptable replacement items. According to a study by Colla and Lapoule [79] in France, the friendliness of the staff was found to be a factor in SQ when it comes to click-and-drive services. Researchers [28] have emphasized that e-retailer quality is largely influenced by the quality of their website. Another study [80] has recommended that enhancing the quality of the retailer’s website can lead to better fulfillment of customers’ needs. Service providers can provide top-notch services by evaluating their website’s quality based on the customers’ perspective. In understanding customer satisfaction, Boyer and Hult examined the effect of the consumer perception of SQ on Buying Intention (BI). According to Zhu and Semeijn’s [81] research, SQ has a favorable connection to the BI of customers. The authors contend that a powerful association exists between the quality of service and the acceptance of online buying [82]. By studying the SQ of websites, researchers indicated that customers’ loyalty to a service provider is mostly determined by the quality of its service. In view of the aforesaid, we postulate that:

Hypothesis (H11). 

SQ positively affects ESI.

2.3.2. Personal Innovativeness (PI)

The level to which a person is eager to endure the existence of novel technology is the primary factor that determines how quickly they adapt to it. This study defines user innovativeness as the willingness to try e-shopping services and new technology. Previous studies [83,84] have shown a link between user creativity and acceptance of technology. Similarly, Leckie et al. [85] showed how service innovation affects total customer engagement, loyalty, and service value appraisal. O’cass and Carlson [86] found that a customer’s perception of how innovative a website is can accurately predict their level of confidence in that website. According to these authors, web users are more likely to believe in a website’s innovation if it has a strong sense of novelty or utility. Online shopping is, therefore, more readily adopted by shoppers who see it as cutting-edge technology. Based on the foregoing, we advance the hypotheses below:

Hypothesis (H12). 

PI is positively linked to ESI.

Hypothesis (H13). 

PI is positively linked with trust.

2.3.3. Perceived Risk (PR)

Perceived risk refers to the assessment that users make when deciding whether to adopt online comments or information on social media [87]. Consumers may feel like they are taking a risk when they shop online because they do not know if a wrong item will be delivered or if they can return or exchange items like they can in a physical store. Concerns about buying could be triggered by a disparity between the actual quality of the item being purchased and the customer’s perception of that item’s quality [32]. In Australia, Mortimer, and his colleagues [25] found that even the most frequent online shoppers are concerned about the possibility of fraud. However, a substantial negative association was observed between PR and the repurchase intention of infrequent e-shoppers. Studies conducted in Denmark and Sweden [23], as well as India [88], evidenced that public relations have a detrimental impact. For Australia, China, Denmark, and Mauritius, the findings [60,89,90,91] show no impact of PR on BI. The notion of PR has not been adequately investigated within the setting of e-shopping in many e-commerce studies [25]. As a result, we incorporate Hansen’s definition of PR [22] into our model and propose the following hypothesis:

Hypothesis (H14). 

PR negatively affects ESI.

2.3.4. Trust

The act of placing one’s reliance or confidence in a specific good or service is known as trust. People will likely buy online if they have faith not just in the capability of the service to fulfill what it advertises but also in the provider of the service. Trust in the context of e-shopping entails customers’ confidence in buying online. Specifically, it encompasses customers’ confidence in online payment security and the safety of their personal information shared on the e-buying websites. Obeidat et al. [92] observed that a positive relationship exists between message trustworthiness and the adoption of an online platform. Most research also identified a positive correlation between trust and intention; however, some have found that the relationship is not as strong as previously thought [93,94,95]. Mobile technology adoption is heavily influenced by trust, according to several studies [95,96,97]. Similarly, researchers [45,93] found that trust has a significant effect on students’ BIs when it comes to using mobile learning and IoT in agriculture. Yet, Kabra et al. [86] discovered that trust and BI are not linked. Jiang et al. [98] found that a favorable link persists between trust in technological advancements and the acceptance of autonomous vehicles. It was also found that helping behavior is related to both affective and cognitive trust [99]. In a similar vein, Algumzi [100] discovered that trust is a factor in e-health utilization. In view of the aforementioned, we propose the hypotheses below:

Hypothesis (H15). 

The connection between trust and ESI is positive.

Hypothesis (H16). 

Trust has a significant and positive relationship with ESB.

2.4. Mediation of Effort Expectancy

The level of expected effort involved in online purchasing can serve as a mediator in the connection that exists between SQ and online shopping intention. Customers believe that buying online is less time consuming than buying at the store. Customers can choose from a wide variety of products in online stores, which makes it easier for them to make decisions. In an online setting, customers have the option to peruse information regarding products and make comparisons between similar products from various brands [101]. By simply navigating a well-designed website, customers can save time and avoid cognitive stress associated with shopping on-site [102]. This is true for all customers, but it is especially important for those who expect to save time and effort by doing their shopping only online. As a result, EE can serve as a bridge between ESI and Online SQ. Amjad-ur-Rehman et al. [20] proved that EE positively mediates the association between SQ and the ESI of consumers. Hariguna [103] also discovered that EE mediates between eGov quality and public behavior (PBIG). This motivates our advancement of the following hypothesis:

Hypothesis (H17). 

Effort expectancy mediates the association between SQ and ESI.

3. Methodology

3.1. Research Design

This is a quantitative and empirical investigation in which an original survey was given to consumers to assess what influences their decision to shop online. Particularly, we used a cross-sectional survey, which involves making an inference about a population at a particular time period.

3.2. Sample and Procedure

This research, which was conducted in Bangladesh, has the potential to endow more global perspectives, since the majority of the previously published materials on this topic originated from affluent countries in the west. Bangladesh is a developing nation in South Asia that has a significant foothold in the e-shopping sector. Hence, e-shoppers throughout Bangladesh (i.e., the target demographic) were contacted to participate in the large-scale administrative survey. Prior to that, a pretest was conducted with the help of 6 relevant scholars (colleagues), and a pilot study involving thirty online shoppers in Bangladesh was also performed and evaluated using Cronbach’s alpha test. The result was 0.905, which showed that the survey is fit to be deployed for a large-scale study.

The cross-sectional survey was administered online during the period August–September 2022, through different e-commerce sites, including Meena clicks, Chaldal, Direct Fresh, and Khas food. To restrict our respondents to those who meet our inclusion criteria in terms of age, employment, e-shopping experience, etc., a “purposive sampling” technique was employed by setting up a group on Messenger and WhatsApp through which a link to the Google Docs survey was sent to each respondent after confirming their eligibility. Willing respondents were briefed about the context, goals, and methods of the study and were also informed that the data collected would be employed solely for academic purposes and that their identities would remain confidential. In addition, information that could be exploited for discriminatory purposes, such as names, races, and religions, was not included in the questionnaire. The survey respected all respondents regardless of their socioeconomic status, as none of their photographs, audio, or video were displayed in the study. We did not express any particular opinions or make any assertions; all we did was display the processed data.

To ensure our sample size was sufficient, we used the G*power tool to determine the minimum required sample size [104]. Based on this calculation, we then determined the number of respondents needed for our study. Cohen’s [105] recommendations were also taken into consideration, which suggest a sample size of at least 184 when dealing with twelve independent constructs or predictors, with an effect size of f² = 0.15, error type 1 of =0.05 and error type 2 of ß = 0.20. As per past literature, a sample size of over 200 is adequate, and a response rate of more than 25 percent is deemed appropriate [106]. In total, 930 Bangladeshis who purchased online were asked to take part in the survey, but 470 of them participated, indicating around 50.5 percent response rate. At the first screening, the responses of 50 participants were eliminated due to missing data and repeated or odd responses. In the end, a valid sample size of 356 was retained.

3.3. Demographic Profile

Table 2. Demographic profile.

Demographics	Classification	Percentages
Gender (out of 356)	Male	65.3
	Female	34.7
Age (Years)	<20	33.2
	20–30	51.8
	30–40	15
Educational Level	Undergraduate	35.5
	Graduate	55.5
	Post graduate	9.0
Online shopping experience	<6 months	11
	6 months to 1 year	20.6
	1 year–2 years	43.4
	>2 years	25

shows that the majority (65.3%) of respondents were male and that 33.2% were below 20 years of age, 51.8% were in the age range of 20–30 years, and 15.0% were between 30 and 40 years. The education level of the participants was recorded and analyzed: 35.5% of the survey respondents held an undergraduate degree, 55.5% possessed a graduate (Master’s degree), and the remaining 9.0% had a doctoral degree. The table further reveals that 11% of the participants had less than six months of online shopping experience, 20.6 percent had 6 to 12 months of experience, 43.4% had 1–2 years of experience, and 25% had more than two years of experience.

3.4. Ethics Statement

According to local and institutional regulations, this study on human subjects does not require an ethics review or permission. However, participants signed a written agreement to participate in the study.

3.5. Measures

During the survey, respondents were requested to evaluate their degree of concurrence with each assertion on a five-point Likert scale, with one denoting “strongly disagree” and five representing “strongly agree.” However, for the e-shopping intention (ESI) measure, its five-point scale ranges from one (every two years) to five (every week).

Table 3. Reliability and Normality Test.

Constructs	Item loading	Mean	SD	Skewness	Kurtosis	Alpha α	CR	AVE
Facilitating Condition (FC) [28]		2.862	0.892	0.148	−0.764	0.817	0.820	0.600
FC1: I possess the necessary resources to participate in online commerce.	0.778
FC2: I am sufficiently knowledgeable about shopping online.	0.722
FC3: My devices support e-shopping (e.g., Facebook).	0.825
Personal Innovativeness (PI) [107].		3.731	0.826	−0.716	0.210	0.880	0.880	0.720
PI1: I like trying out new things.	0.746
PI2: I wouldn’t think twice about using technology to buy something online.	0.901
PI3: Among the people in my network, I am typically the one who is the first to test out new technology.	0.891
Performance Expectancy (PE) [28]		2.729	1.080	0.337	−1.049	0.824	0.830	0.610
PE1: E-commerce helps me find things quickly.	0.804
PE2: I can find some things and services online that are harder to find in stores.	0.717
PE3: I can boost my productivity by using e-shopping.	0.826
Trust (TT) [97]		4.319	0.719	−1.722	3.798	0.869	0.870	0.690
TT1: I believe that buying online is trustworthy.	0.839
TT2: I feel secure making purchases online since my private information will be protected.	0.893
TT3: I believe that payment online in e-shopping is safe.	0.762
Hedonic Motivation (HM) [38]		3.273	0.945	−0.280	−0.833	0.768	0.770	0.540
HM1: Buying through e-stores is fun.	0.609
HM2: Buying through e-stores is enjoyable.	0.853
HM3: E-stores are visually appealing and attractive.	0.715
Price Value (PV) [38]		3.533	0.883	−0.473	−0.534	0.850	0.850	0.660
PV1: Products in E-store are reasonably priced	0.767
PV2: Online stores offer good value.	0.809
PV2: At the existing price, the e-store supplies a better value.	0.851
Social Influence (SI) [107]		3.183	1.053	0.018	−1.206	0.906	0.910	0.770
SI1: The individuals who hold significance in my life believe that I should make purchases from online stores.	0.878
SI2: People who have influence over my actions think that I ought to buy products from online merchants.	0.864
SI3: People I care about tell me that I should shop online.	0.883
Service Quality (SQ) [32]		2.209	0.978	0.856	−0.244	0.890	0.900	0.740
SQ1. My online store’s staff is helpful.	0.847
SQ2. The check-out process at my online store is quick.	0.806
SQ3. The services provided by my online store are sufficient (i.e., picking the orders, payment, etc.)	0.929
Effort Expectancy (EE) [28]		2.318	0.894	0.735	−0.235	0.866	0.870	0.680
EE1: E-shopping is a simple process that anyone can pick up quickly.	0.844
EE2: I find it easy to shop online.	0.796
EE3: E-shopping is easy to master.	0.842
Perceived Risk (PR) [32]		2.223	0.822	0.796	0.183	0.796	0.800	0.570
PR1: It is more difficult to return or exchange items in an e-store than in a traditional retail setting.	0.783
PR2: When purchasing food online, the danger of receiving the wrong item is a concern.	0.650
PR3: I worry about the e-store goods’ quality.	0.826
Habit (HB) [37]		3.593	0.911	−0.557	−0.548	0.853	0.860	0.660
HB1: I’ve gotten into the habit of using shopping apps.	0.880
HB2: I’m used to ordering online.	0.740
HB3: Shopping apps are my addiction.	0.820
E-shopping Intention (ESI) [37]		3.503	0.927	−0.625	−0.474	0.841	0.830	0.620
ESI1: I like shopping apps, and will continue to use them.	0.752
ESI2: When it comes to purchasing, I’m planning to keep using apps in the future.	0.774
ESI3: Going forward, I anticipate using shopping applications on a regular basis.	0.834
E-Shopping Behavior (ESB) [37]		3.960	0.730	−1.346	1.166	0.866	0.870	0.690
ESB1: I use shopping apps sometimes.	0.954
ESB2: I shop with smartphone apps often.	0.843
ESB3: I always shop via mobile apps.	0.679

presents measurements adopted from several studies. The items of ESI, ESB, and the habit, as well as those of FC, PE, and EE, were adopted from past researchers, such as Tak and Panwar [37]. The measurement items of PI and SI were adopted from Alkawsi et al. [107], while those of HM and PV were culled from Alam et al. [38]. The items for measuring PR and SQ were taken from Van Droogenbroeck and Van Hove [32], and items for Trust were adapted from Chao [97].

3.6. Data Analysis Methods

This study used a two-stage multi-analytical technique, integrating SEM with the ANN. The proposed hypotheses were tested with the SEM. The SEM method is appropriate for this research since it seeks to validate the sufficiency of the UTAUT-2 model in terms of comprehending the behaviors associated with e-shopping. In addition to this, SEM corrects any measuring inaccuracies that may occur in the items being measured [108]. Moreover, SEM looks at all the relationships that depend on each other in a single analysis [108] and usually gives results that are free of mistakes. In this confirmatory research, the SEM approach relied on covariance to examine the relationships among variables [109]. The data was analyzed using IBM-SPSS version 26 and AMOS version 26. The SEM analysis was then used to inform the input neurons for the ANN model, following previous literature [110]. The decision to use ANN was driven by the presence of non-normal data distribution and non-linear relationships between the independent and dependent variables. ANN also proves to be robust to disturbances, outliers, and datasets with limited sample sizes. The ANN analysis was conducted using the neural network module of IBM’s SPSS.

4. Results

4.1. Data Screening and Normality

The data collected from the self-administered questionnaire underwent a data screening process. We first analyzed the missing data, outliers, normality, and correlations among items. Then, we computed the covariances and Cronbach’s alpha to assess the fitting of each construct. Finally, the Confirmatory Factor Analysis (CFA) was performed for each construct.

Prior to data processing, it is a common practice to screen the data for any inaccuracies. As a result, preliminary data screening was carried out. Due to the pattern of online data collection, there were no errors or omissions in the latest 325 records. Hence, the mean of all the data points, excluding those that were statistically significant outliers, was chosen. In line with past literature [111,112], the values of skewness and kurtosis of the variables fell within the acceptable range (one to three).

To check for multivariate normality, this study used the Web Power online tool [113], and the results revealed that p-values for Mardia’s multivariate skewness and kurtosis were over 0.05, indicating the normality. Thus, to accommodate normal data, the current study utilized covariance-based SEM. Furthermore, the tolerance level and Variance Inflation Factors (VIFs) were employed to assess whether or not the independent variables were affected by multicollinearity [114]. Accordingly, the tolerance values, as a measure of VIF, ranged from 0.245 to 0.967, raising no concern about multicollinearity. Furthermore,

Table A1. Test of linearity.

Particular			Sum of Squares	df	Mean Square	F	Sig.
ESI ∗ TT	Between Groups	(Combined)	34.114	11	3.101	3.881	0.000
		Linearity	13.487	1	13.487	16.880	0.000
		Deviation from Linearity	20.627	10	2.063	2.582	0.005
	Within Groups		325.993	408	0.799
ESI ∗ PI	Between Groups	(Combined)	55.924	12	4.660	6.236	0.000
		Linearity	39.935	1	39.935	53.434	0.000
		Deviation from Linearity	15.988	11	1.453	1.945	0.033
	Within Groups		304.183	407	0.747
ESI ∗ PV	Between Groups	(Combined)	28.939	11	2.631	3.241	0.000
		Linearity	19.253	1	19.253	23.720	0.000
		Deviation from Linearity	9.686	10	0.969	1.193	0.293
	Within Groups		331.168	408	0.812
ESI ∗ EE	Between Groups	(Combined)	20.902	11	1.900	2.286	0.010
		Linearity	6.193	1	6.193	7.449	0.007
		Deviation from Linearity	14.709	10	1.471	1.769	0.064
	Within Groups		339.205	408	0.831
ESI ∗ FC	Between Groups	(Combined)	39.707	12	3.309	4.203	0.000
		Linearity	9.352	1	9.352	11.880	0.001
		Deviation from Linearity	30.355	11	2.760	3.505	0.000
	Within Groups		320.400	407	0.787
ESI ∗ HM	Between Groups	(Combined)	15.884	11	1.444	1.711	0.069
		Linearity	10.814	1	10.814	12.818	0.000
		Deviation from Linearity	5.070	10	0.507	0.601	0.813
	Within Groups		344.223	408	0.844
ESI ∗ HB	Between Groups	(Combined)	50.485	12	4.207	5.530	0.000
		Linearity	35.190	1	35.190	46.257	0.000
		Deviation from Linearity	15.296	11	1.391	1.828	0.048
	Within Groups		309.621	407	0.761
ESI ∗ SQ	Between Groups	(Combined)	21.787	11	1.981	2.389	0.007
		Linearity	8.851	1	8.851	10.675	0.001
		Deviation from Linearity	12.935	10	1.294	1.560	0.116
	Within Groups		338.320	408	0.829
ESI ∗ PE	Between Groups	(Combined)	42.875	12	3.573	4.584	0.000
		Linearity	15.975	1	15.975	20.496	0.000
		Deviation from Linearity	26.899	11	2.445	3.137	0.000
	Within Groups		317.232	407	0.779
ESI ∗ PR	Between Groups	(Combined)	44.207	11	4.019	5.191	0.000
		Linearity	25.655	1	25.655	33.135	0.000
		Deviation from Linearity	18.552	10	1.855	2.396	0.009
	Within Groups		315.900	408	0.774
ANOVA for ESB
ESB ∗ TT	Between Groups	(Combined)	24.508	11	2.228	4.577	0.000
		Linearity	10.790	1	10.790	22.167	0.000
		Deviation from Linearity	13.718	10	1.372	2.818	0.002
	Within Groups		198.608	408	0.487
ESB ∗ ESI	Between Groups	(Combined)	42.720	12	3.560	8.032	0.000
		Linearity	23.361	1	23.361	52.706	0.000
		Deviation from Linearity	19.359	11	1.760	3.971	0.000
	Within Groups		180.396	407	0.443
ESB ∗ HB	Between Groups	(Combined)	16.952	12	1.413	2.789	0.001
		Linearity	11.056	1	11.056	21.826	0.000
		Deviation from Linearity	5.896	11	0.536	1.058	0.394
	Within Groups		206.164	407	0.507

Note: FC = facilitating condition, PI = personal innovativeness, PE = performance expectancy, TT = trust, HM = hedonic motivation, PV = price value, SQ = service quality, EE = effort expectancy, PR=perceived risk, HB = Habit, ESI = shopping intention, ESB = shopping behavior.

supports the linear and non-linear links existent between the independent and dependent variables.

4.2. Reliability and Common Method Bias

Due to the cross-sectional nature of the investigation, there is a possibility of Common Method Bias (CMB)—a spurious variation in measuring the constructs—which could negatively influence the validity of scientific findings [115]. Harman’s Single Factor test is commonly used to quantify CMB in self-reported data [116]. Using SPSS, unrotated Exploratory Factor Analysis (EFA) was deployed to find the number of factors that account for the variance in variables, and only 19% of the variance was explained by a single factor. As a benchmark, the amount to which items are associated is considered to be the variance. Hair et al. [109] stipulated that standardized loading estimates ought to be at least 0.5, and all loading values were above 0.5 and significant at the 5% level. In addition to Harman’s single-factor test, we also used the common latent factor approach. We took an unmeasured latent factor to the measurement model and related it to each observed item. Examining the resulting model’s path coefficients showed a typical variance of 28.5%, which is less than the 50% threshold [117].

Table 3 shows that the Average Extracted Variance (AVE) values were more than 0.5 [118], and the CR score was greater than the benchmark of 0.7, indicating that the model is good and can be used in preliminary research [119]. The study’s designs are statistically acceptable based on the stated criteria.

4.3. Validity

Discriminant validity estimates the degree to which individual items exhibit singularity with the respective constructs being evaluated, and it is popularly measured using the Fornell–Larcker criterion. The criterion has drawn severe criticism from scholars who indicated that it is unsuitable for evaluating discriminant validity in normal research circumstances. As a result, Hetero Trait and Mono Trait (HTMT), a multitrait–multimethod matrix-based technique, was applied to assess the discriminant validity. Furthermore, the study investigated HTMT’s robustness due to its dominance over Fornell–Larcker in many scenarios [120]. Since the HTMT values were smaller than 0.85 [120], the discriminant validity was proved (

Table 4. Discriminant Validity (Fornell Larker Criterion and HTMT), VIF and R².

	FC	PI	PE	TT	HM	PV	SI	SQ	EE	PR	HB	ESI	ESB
Fornell Larcker criterion
FC	0.775
PI	0.048	0.849
PE	0.065	0.140	0.781
TT	0.006	0.217	−0.133	0.831
HM	−0.028	0.075	0.015	0.151	0.735
PV	−0.073	0.152	0.112	0.203	0.178	0.812
SI	0.084	0.035	0.011	0.101	−0.037	0.026	0.877
SQ	0.043	0.038	0.039	−0.084	0.043	−0.015	0.037	0.860
EE	0.047	−0.199	−0.128	−0.036	0.016	−0.084	0.093	0.240	0.825
PR	−0.036	−0.323	−0.165	−0.126	−0.026	−0.089	0.078	0.031	0.707	0.755
HB	−0.068	0.464	0.148	0.192	0.181	0.149	−0.007	0.022	−0.293	−0.374	0.812
ESI	.161	0.333	0.211	0.194	0.173	0.231	0.026	0.157	−0.131	−0.267	0.313	0.787
ESB	0.025	0.250	0.085	0.220	−0.011	0.126	−0.010	0.093	−0.082	−0.151	0.223	0.324	0.831
Hetero Trait and Mono Trait (HTMT)
FC
PI	0.053
PE	0.072	0.160
TT	0.017	0.247	−0.160
HM	−0.037	0.098	0.037	0.170
PV	−0.086	0.176	0.140	0.243	0.197
SI	0.092	0.051	0.035	0.150	−0.062	0.049
SQ	0.054	0.056	0.053	−0.102	0.068	−0.028	0.057
EE	0.062	−0.224	−0.148	−0.055	0.029	−0.092	0.124	0.255
PR	−0.074	−0.351	−0.183	−0.140	−0.037	−0.098	0.098	0.046	0.726
HB	−0.077	0.483	0.157	0.226	0.198	0.170	−0.027	0.058	−0.326	−0.398
ESI	0.178	0.353	0.226	0.234	0.197	0.255	0.051	0.166	−0.153	−0.326	0.338
ESB	0.038	0.268	0.099	0.245	−0.032	0.160	−0.018	0.110	−0.108	−0.166	0.241	0.339
VIF	1.312	1.414	1.125	1.217	1.093	1.137	1.035	1.222	1.933	2.075	1.474	1.368
Tolerance	0.925	0.707	0.889	0.821	0.915	0.879	0.967	0.818	0.254	0.245	0.679	0.731
R2				0.060					0.070			0.300	0.150

Note: The square root of AVE is indicated by the bold elements. FC = facilitating condition, PI = personal innovativeness, performance expectancy, TT = trust, HM = hedonic motivation, SQ = shopping quality, PV = price value, EE = effort expectancy, ESI = shopping intention, SI = social influence, ESB = shopping behavior, PR = perceived risk, and HB = Habit.

).

The structural model of the study is depicted in Figure 2 and

Table 5. Structural Model and Hypothesis Testing Result.

Hypotheses	STD Beta	STD Error	t-Values	p-Values	Significance (p < 0.05)
H1: FC→ESI	0.178	0.051	3.351 ***	0.000	Accepted
H2: EE→ESI	0.128	0.049	2.408 **	0.016	Accepted
H3: SI→ESI	−0.006	0.038	−0.127	0.899	Rejected
H4: PE→ESI	0.186	0.040	3.506 ***	0.000	Accepted
H5: PE→ESB	0.010	0.042	0.190	0.850	Rejected
H6: HM→ESI	0.134	0.052	2.517 **	0.012	Accepted
H7: PV→ESI	0.163	0.049	3.135 ***	0.002	Accepted
H8: HB→ESI	0.160	0.051	3.100 ***	0.002	Accepted
H9: HB→ESB	0.116	0.053	2.233 **	0.026	Accepted
H10: ESI→ESB	0.290	0.058	5.156 ***	0.000	Accepted
H11: SQ →ESI	0.135	0.038	2.599 ***	0.009	Accepted
H12: PI→ESI	0.161	0.059	3.079 ***	0.002	Accepted
H13: PI→TT	0.235	0.049	4.245 ***	0.000	Accepted
H14: PR→ESI	−0.271	0.056	−4.917 ***	0.000	Accepted
H15: TT→ESI	0.114	0.067	2.176 **	0.030	Accepted
H16: TT→ESB	0.174	0.069	3.370 ***	0.000	Accepted
H17: SQ→EE→ESI	0.033	0.014	2.386 **	0.017	Accepted

Note: ** Significant at 5% level, *** Significant at 1% level, FC = facilitating condition, PI = personal innovativeness, PE = performance expectancy, TT = trust, HM = hedonic motivation, EE = effort expectancy, PV = price value, SQ = service quality, SI = social influence, PR = perceived risk, HB = Habit, ESI = shopping intention, and ESB = shopping behavior.

. The measurement model was first subjected to a CFA test, after which the structural model was validated. During this process, the proposed model’s goodness of fit indices was assessed. The findings of the SEM showed that there was a good match between the data (χ2/df = 2.227). Since the actual result was 0.043, it may be concluded that the RMSEA threshold value of less than 0.08 was successfully passed [121]. A number of fit indices (GFI, CFI, TLI, IFI, etc.) met the 0.9 or higher requirement [122].

4.4. Structural Model

In this study, the significance of the path coefficient, t-value, and p-value, as well as R² (the percentage of variance that was explained), were utilized to investigate the structural model and the hypotheses. With a significance level of 5%, the critical value for both tests was 1.96. Our findings revealed that the R² value for ESI and ESB was 0.300 and 0.150, respectively. Additionally, the values for EE and trust were observed to be 0.070 and 0.060, respectively, which are all considered to be moderate [109]. Prior to the SEM, the path analysis, which suggests that the variance in the endogenous variable can be explained by the exogenous variables directly or indirectly, was performed. The outcome of the path analysis indicates that the model was an excellent match (χ2/df = 2.242, IFI = 0.933, TLI = 0.929, CFI = 0.913, and RMSEA = 0.055) [123]. Table 5 shows the direct effects and their relevance.

The result revealed that exogenous variables—FC (beta = 0.178; t = 3.351; p < 0.01), PI (beta = 0.161; t = 3.079; p < 0.01), PE (beta = 0.235; t = 4.245; p < 0.01), HM (beta = 0.134; t = 2.517; p < 0.01), trust (beta = 0.114; t = 2.176; p < 0.01), PV (beta = 0.163; t = 3.135; p < 0.01), SQ (Beta = 0.135; t = 2.599; p < 0.01), EE (beta = 0.128; t = 2.408; p < 0.01), PR (beta = −0.271; t = −4.917; p < 0.01), and habit (beta = 0.160; t = 3.100; p < 0.01)—were significantly associated with the ESI. The construct SI (β = −0.006; t = −0.127; p > 0.01) was not found to be related to ESI. Moreover, habit (β = 0.116; t = 2.233; p < 0.01), ESI (β = 0.290; t = 5.156; p < 0.01), and trust (β = 0.174; t = 3.370; p < 0.01) were observed to be significantly linked to ESB, while PE was not significantly related to ESB (β = 0.010; t = 0.190; p > 0.05). Overall, the study found that all exogenous variables (FC, PI, PE, HM, trust, PV, SQ, EE, PR, and habit) were significantly associated with the ESI except for SI, and habit, ESI and trust were significantly linked to ESB (except PE), supporting H1, H2, H4, and H6–H16, but invalidating H3 and H5.

4.5. Mediation Effect of Effort Expectancy Constructs

The Sobel test was utilized in the current research to determine the potential mediation role of EE in the connection between SQ and the ESI, as proposed by scholars [124]. This test has been favored over the bootstrapping method because the data were regularly distributed. The proper analysis to assess the joint significance of the indirect effect is the indirect effect method, which can furnish the necessary support. The test showed that attitude (beta = 0.033, t value = 2.386, and p value < 0.05) mediates the relationship between SQ and ESI, thus validating hypotheses 17, 18, and 19 (Table 4).

4.6. ANN Analysis

In line with a prior study [110], this investigation went one step further and used the relevant variables that emerged from the path analysis of SEM-PLS as the input neurons for the neural network model. The utilization of ANN is necessary due to the non-normal distribution of data and the non-linear relationship between independent and dependent variables. ANN analysis is robust against interference, anomalies, and limited sample sizes. The study employed the neural network module of IBM’s SPSS in performing the ANN examination. The ANN method does not necessitate that the data follow a normal distribution and can identify both linear and non-linear correlations [125]. Typically, an ANN will have one input level, one hidden unit level and one output level [126]. Two separate ANN models were developed, one for each set of dependent variables (research model output neurons) (see Figure 3). Figure 3 shows that there were ten key influencing factors for ESI (FC, EE, PI, SQ, TT, HM, PV, PE, PR, and HB) and three for ESB (ESI, TT, and HB).

Additionally, to avoid over-fitting issues in the ANN model, as indicated by the Root Mean Square Error (RMSE), a ten-fold cross-validation approach was employed. In this process, 90% of the data was designated for training, while the rest of the 10% was allocated for testing [127].

The RMSE values for Models A and B are presented in

Table A2. RMSE values for Model A in a ten-fold ANN.

Networks	Training			Testing			Total Samples
	N	SSE	RMSE	N	SSE	RMSE
ANN1	368	141.196	0.619	52	20.218	0.624	420
ANN2	379	144.855	0.618	41	13.733	0.579	420
ANN3	374	129.038	0.587	46	19.567	0.652	420
ANN4	378	133.767	0.595	42	11.053	0.513	420
ANN5	372	159.352	0.654	48	12.832	0.517	420
ANN6	382	141.440	0.608	38	10.311	0.521	420
ANN7	377	116.101	0.555	43	11.361	0.514	420
ANN8	375	129.655	0.588	45	19.577	0.660	420
ANN9	372	133.466	0.599	48	13.632	0.533	420
ANN10	371	165.168	0.667	49	10.976	0.473	420
Mean		139.404	0.609	Mean	14.326	0.559
SD		14.573	0.033	SD	3.939	0.066

and

Table A3. RMSE values for Model B in a ten-fold ANN.

Networks	Training			Testing			Total Samples
	N	SSE	RMSE	N	SSE	RMSE
ANN1	366	150.507	0.641	54	15.790	0.541	420
ANN2	385	163.352	0.651	35	8.082	0.481	420
ANN3	368	174.997	0.690	52	8.468	0.404	420
ANN4	377	152.868	0.637	43	18.983	0.664	420
ANN5	372	151.512	0.638	48	19.274	0.634	420
ANN6	376	161.579	0.656	44	14.105	0.566	420
ANN7	378	161.927	0.655	42	13.635	0.570	420
ANN8	386	162.803	0.649	34	17.390	0.715	420
ANN9	385	158.165	0.641	35	10.242	0.541	420
ANN10	389	175.102	0.671	31	16.000	0.718	420
Mean		161.281	0.653	Mean	14.197	0.583
SD		8.681	0.016	SD	4.094	0.101

in the Appendix A. The minimum value of RMSE was 0, while there was no upper limit. Model A had RMSE values of 0.609 for training and 0.559 for testing, and Model B showed RMSE values of 0.653 for training and 0.583 for testing. These results suggest that the models have a strong fit [127].

Furthermore, we conducted a sensitivity analysis to calculate the influence of each input neuron on the dependent variable (refer to

Table 6. Sensitivity Analysis for Models A and B.

For Model A
	NN1	NN2	NN3	NN4	NN5	NN6	NN7	NN8	NN9	NN10	AI	NI (%)
TT	0.535	1.000	0.500	0.154	0.640	0.691	0.467	0.229	0.528	1.000	0.574	0.713
PI	0.931	0.556	1.000	0.493	0.756	1.000	0.701	0.559	0.740	0.898	0.763	0.947
EE	0.406	0.478	0.501	0.748	0.195	0.275	0.710	0.523	0.509	0.326	0.467	0.580
SQ	0.449	0.541	0.550	0.327	0.305	0.342	0.415	0.352	0.399	0.186	0.387	0.480
FC	0.925	0.714	0.484	0.454	0.335	0.693	0.491	0.357	0.759	0.327	0.554	0.687
PV	0.998	0.788	0.655	0.389	0.220	0.675	0.593	0.458	0.530	0.699	0.600	0.745
HM	0.113	0.359	0.216	0.219	0.327	0.189	0.384	0.234	0.213	0.462	0.272	0.337
PE	0.493	0.584	0.256	0.439	0.245	0.557	0.355	0.385	0.396	0.262	0.397	0.493
HB	1.000	0.653	0.620	0.535	1.000	0.367	0.443	0.319	0.582	0.234	0.575	0.714
PR	0.980	0.596	0.946	1.000	0.583	0.771	1.000	1.000	1.000	0.186	0.806	1.000
For Model B
ESI	1.000	1.000	0.556	1.000	1.000	0.966	1.000	1.000	0.925	1.000	0.945	1.000
TT	0.293	0.802	1.000	0.664	0.590	1.000	0.859	0.373	1.000	0.257	0.684	0.724
HB	0.395	0.273	0.558	0.681	0.225	0.149	0.243	0.271	0.596	0.392	0.378	0.401

Note: AI = average importance, NI = normalized importance, FC = facilitating condition, PI = personal innovativeness, HB = Habit, TT = trust, HM = hedonic motivation, EE = effort expectancy, ESB = shopping behavior, PE = performance expectancy, PR = perceived risk, PV = price value, ESI = shopping intention, and SQ = service quality.

). The normalized significance of each predictor was assessed by dividing its mean rank by the maximum mean rank of the predictors [128]. The results revealed that the PR had the greatest impact on predicting the ESI for Model A, followed by PI, PV, HB, TT, FC, PE, SQ, and HM. Meanwhile, for Model B, ESB was the most crucial factor in determining ESI, with TT and HB being the next significant factors.

5. Discussion

The study determined the predictors of e-shopping behavior among customers in Bangladesh using the UTAUT 2 model, and the outcome revealed that all the proposed predictors of usage intention and behavior hold except for SI and PE, which were not statistically significant. The most significant of the factors were perceived risk (β = −0.271) for e-shopping intention and e-shopping (β = 0.290), and for e-shopping behavior.

Hypothesis 1 postulated that FC affects e-shopping intention, and the present study confirmed the relationship. This signifies that people with the necessary facilities are more inclined to e-shopping, which is in line with the findings of past studies [38,39]. Hypothesis (H2) predicted that EE is connected to the ESI, which was validated in this study. The result is according to the past research of Shaikh et al. [41] and Quaosar et al. [44]. This suggests that people are predisposed to technology whose usage requires less effort. In other words, easy access, operation, and feedback systems influence people’s use of a system.

According to H3, social influence is associated with e-shopping intention. In other words, external endorsement or recommendation may influence the decision pattern of an individual. However, this study did not support the proposition, as there was no significant relationship between SI and ESI, which contradicts the findings of several other studies [38,47,49]. This may be attributed to the increasing reliance of people on social or other media for shopping decisions rather than on peers or friends.

As indicated in H4, performance expectancy was found to predict e-shopping intention. Through e-shopping, people can save time, resulting in greater productivity. This outcome is supported by earlier studies [54], which concluded that the greater the usefulness of a system, the greater the customer’s usage intent towards it. In contrast, the relationship between performance expectancy and e-shopping behavior, as proposed in H5, was not significant. This result is similar to the original UTAUT model, which excluded this relationship [26], signifying the possible role of mediating or moderating factors in the intention–behavior relationship. E-commerce businesses need to ensure that their online shopping systems are easy to use and provide a positive user experience. This can be achieved by designing intuitive and user-friendly interfaces, providing accurate and relevant product information and offering a range of payment and delivery options.

According to the propositions (H6–H8), HM, PV, and habit are linked to the e-shopping intention. The outcome of this study validated the hypotheses and agrees with the findings of the past research [60], which identified that the perceived enjoyment of system use and the value for prices and consumers’ habits of regular use in a system significantly influence usage intention. In other words, the greater the HM, PV, and habit, the greater the chance of e-shopping intention. The more an individual is motivated by pleasure, enjoyment, or entertainment, the more likely they are to intend to engage in e-shopping. The more an individual values the price of a product or service, the more likely they are to intend to engage e-shopping. The more an individual has developed a habit of e-shopping, the more likely they are to intend to engage in e-shopping.

Similarly, the proposed H9 of a connection between habit and e-shopping behavior was significant in Bangladesh. This outcome aligns with the results of past studies [68,129], suggesting that consumers in Bangladesh habitually make use of online shopping. In e-shopping, the habit could be the tendency to check e-commerce websites regularly or the preference to buy from certain e-commerce websites. This result suggests that habit plays a significant role in shaping an individual’s e-shopping behavior.

As expected, the result showed that e-shopping intention is strongly connected with e-shopping behavior (H10). This outcome supports the previous research [28,31,37], indicating that greater consumer intention of e-shopping results in a greater chance of ESB. The quality of e-shopping was observed to be an important predictor of ESI, thus validating the H11 and corroborating the previous research [82]. This indicates that if the e-shopping platform provides clear information, greater options of choice, an easy payment gateway, and return facilities, customers’ interest in online purchases will increase.

Personal innovativeness is another factor that predicts e-shopping intention (H12). People who are innovative or adopt any technology earlier than others are more likely to try out novel technology or system. In the current study, we found a strong bond between personal innovativeness and e-shopping intention, thus corroborating the outcomes of previous research [83,84]. Similarly, past research [86] identified the presence of a link between personal innovativeness and consumer trust (H13), which has been confirmed in our study, as the relationship was found to be significant. This reveals that the higher the consumer’s innovativeness, the greater their trust in the e-stores. According to H14, perceived risk is a relevant predictor of intention (H14). As per our result, the hypothesis was supported, albeit negatively. In other words, PR is negatively related to ESI, thus supporting the research of Mortimer et al. [25] while contradicting other research works [60,91]. This means a greater perceived risk of e-shopping lowers the intention of consumers to participate in e-shopping and vice versa. Perceived risk in e-commerce refers to the potential negative consequences of engaging in a particular behavior, such as e-shopping. These negative consequences could be financial loss, security breaches, or product quality issues. E-commerce businesses need to take appropriate risk management measures, which involve improving the security of their website, providing transparent and accurate product information and implementing a return policy that ensures consumer satisfaction.

Trust plays a crucial part in promoting e-purchasing behavior, as it translates consumers’ positive expectations and actions into e-shopping patronage for purchases. Confidence has already been found to be a fundamental requirement for e-commerce growth in a number of earlier research [130,131]. Additionally, the literature highlights that a deficiency in trust is the primary factor for the unfavourability towards e-shopping or the abandonment of purchases. Due to the users’ failure to adopt safety measures, the rate of disposal in cyberspace has become extremely high [132]. If the e-store ensures confidence in the order-processing procedures, customers become motivated to buy via e-shopping (H15 and H16). This outcome supports the previous research [97].

The findings also showed that EE mediates the association between SQ and ESI (H17). Online shopping SQ has a direct and considerable effect on ESI. This outcome corroborates with the past studies of Amjad-ur-Rehman et al. [20] and Hariguna [103]. However, when a customer expects a high level of performance and minimal effort, the association between SQ and online shopping becomes stronger. In other words, the relationship between service quality and e-shopping intention is partially explained by the perceived effort required to engage in e-shopping. By improving the ease of use of their e-commerce websites, businesses can increase the perceived effort expectancy, which, in turn, increases e-shopping intention. Similarly, by enhancing their service quality, businesses can increase the overall satisfaction of their customers, which, in turn, increases e-shopping intention.

Furthermore, ANN analysis was used to identify the key influencing factors of ESI and ESB, where ten factors (FC, EE, PI, SQ, TT, HM, PV, PE, PR, and HB) were found to influence ESI and three (ESI, TT, and HB) for ESB. The ANN models demonstrated a strong fit, with low RMSE values for both training and testing sets. Sensitivity analysis was also performed to evaluate the influence of each input neuron on the dependent variables. Overall, the study’s results provide valuable insights into the factors that affect ESI and ESB and highlight the significance of using different statistical techniques for analyzing complex relationships.

6. Implications

6.1. Theoretical Contribution

The study contributes theoretically in terms of model extension and development of new constructs, methodology, and results. First, existing literature, such as Rehman et al. [28] studied the e-shipping behavior based on the UTAUT; however, this research focuses on the use of the UTAUT2 model with further extension to understand the e-shopping behavior. Second, many past studies were conducted from the developed country’s perspectives, with only a few studies emanating from the developing countries, including Bangladesh. Furthermore, this is one of the first studies that combined novel methodologies (SEM-ANN approach) to investigate the predictors affecting ESI and behavior and the mediating effect of effort expectancy, all of which have been mostly overlooked in previous studies.

Regarding the new constructs, the study incorporated some new variables to make the model more comprehensive and context specific. First, the e-shopping SQ was incorporated following Rehman et al. [28], which is a novel consideration in the Bangladeshi context. Second, the study also included ESB in the model. This inclusion covers the limitations of many studies that focus only on intentions. Third, this study considered the role of PR on ESI, thus overcoming the limitations of Cabrera-Sánchez et al. [129]. Fourth, most studies based on the UTAUT model experimented with the moderating variable; here in this study, we offered the mediating variable of effort expectancy, which is a contribution of this research to the shopping literature. With all these new constructs, this study contributes to the e-commerce literature significantly.

Finally, the study established new results in this field. First, the mediation role of effort expectancy was tested and established between the SQ and e-shopping intention. This result corroborates the past result adequately. Second, SI was identified with no effect on ESI, which is in opposition to the original UTAUT model. Likewise, EE was included in this research as a new variable but was found to be irrelevant to the ESB context. This experiment supplies new results, which will contribute to academia seeking necessary explanations for the predictors of ESI.

6.2. Managerial Implications

This research provides various implications for managers. First, the outcome shows that customers place a high value on the hedonic and habitual aspects of a product. This information will guide marketers in making the online purchasing experience more exciting. To better connect and engage their customers, marketers can utilize social networking sites. By enabling customers to share their online-buying experience, word-of-mouth marketing may be generated, thus developing a situation that benefits both parties involved (customers and marketers). With the aid of social media, marketers can advertise special deals or incentives to encourage customers to shop online. Technology-driven innovation may help to improve the user friendliness of the interface for consumers. Easy-to-use interfaces will incentivize people who are not tech savvy to use online stores, thereby improving the profit of marketers.

Second, it is necessary to note that the experience of the COVID-19 pandemic has brought with it a new list of challenges for retailers, such as the need to retain as many existing customers as possible. In our above analysis, “habit” was not only the most actionable aspect but also a strong indicator of the intention to use an e-shopping service during the post-COVID-19 pandemic era. Hence, there is a greater necessity for online store retailers to grow features that assist habit formation, such as the shopping lists discussed previously.

Third, e-shopping service quality was a significant predictor of e-shopping intention. Marketers should lay more emphasis on their support services, which include the reduction of the delivery time of ordered products and assurance of the safety and security of products. In addition, payment gateway improvement is another maintenance service demanded by consumers. Marketers must maintain a friendly delivery service by ensuring their staff is well trained. Fourth, perceived risk is the most dominant predictor of e-shopping behavior, implying that people are concerned about the prospective risk of online shopping. Marketers should address the issue of product quality not matching the provided information on the e-stores, as most of the complaints from customers relate to product quality dissatisfaction. Although e-stores provide an easy return and refund policy, this policy must be strictly regulated. Government and other policymakers must also enact policies to protect consumerism. Currently, the consumer association of Bangladesh (CAB) is playing a praiseworthy role in establishing consumer rights to protect customers. This should be more visible and active with government supports enlarging supervision capability and empowerment.

7. Limitations and Recommendations for Future Studies

Despite the study’s numerous contributions, it is not bereft of limitations. First, in the proposed model, three new variables were included to expand the UTAUT 2 framework and obtain a fragmented opinion of purchasing behavior. Some other variables can also be included, such as technology readiness, customer satisfaction, etc., to make the model more comprehensive. Second, to investigate the possibility of additional effects that had not been taken into account previously, new moderator variables, in addition to those that were included in the first UTAUT, should be investigated. These might enable us to distinguish between different types of consumer behavior and identify potential market segments. In subsequent research, the level of technological preparedness might be investigated as a potential moderator or mediator in an effort to close the intention–behavior disparities. Third, this study applied a cross-sectional research design with convenient sampling, which may have limitations in terms of generalization. Furthermore, the majority of the respondents in this study were male, which may also limit generalizability in different settings. Future research should come up with an experimental research design with other sampling methods, such as systemic sampling with a large data set to cover these limitations.

8. Conclusions

The research intended to determine factors encouraging people to shop online in Bangladesh. The research also investigated whether or not effort expectation plays a mediating function in the connection between SQ and the desire to participate in e-commerce activities. The study employed a well-fitting model, revealing that trust, habit, and e-shopping intention all had favorable effects on shoppers’ behavior when purchasing goods online. Similarly, PI, FC, PE, EE, habit, PR, PV, HM, SQ, and trust were significantly linked with e-shopping intention. On the other hand, the study invalidated the influence of social presence and facilitating conditions on e-service adoption. Additionally, the research showed that the link between SQ and ESB is mediated by EE. Overall, the findings of this study provide valuable insights into the factors that influence customers’ online shopping behavior and can help e-commerce companies better understand their customers and tailor their marketing strategies accordingly. By focusing on the factors identified in this study, e-commerce companies can improve their customers’ shopping experience and increase their sales revenue, ultimately contributing to the growth and success of the e-commerce industry as a whole.